Meta’s work on fairness in personalized ads addresses risks of discrimination in both targeting and delivery. Early changes restricted sensitive targeting options in housing, employment, and credit ads, while tools like the Ad Library increased transparency. The focus has since shifted to ad delivery outcomes, where algorithms may unintentionally skew results. To counter this, Meta developed the Variance Reduction System (VRS), a machine learning method that reduces demographic variance between eligible and actual ad viewers. Using privacy-preserving techniques, VRS adjusts ad delivery without accessing individual demographic data, aiming to improve equity while balancing privacy, fairness, and policy challenges.

Join for free to read